Citation:
Sezgin, Tevfik Metin, and Randall Davis. "HMM-based efficient sketch recognition." Proceedings of the 10th international conference on Intelligent user interfaces. ACM, 2005.
Summary:
This paper deals with the application of HMMs to sketch recognition, that is treating sketches a series of incremental states built upon every stage of interaction. The goal is to capture variations in drawing style for various figures, and break these down into a sequence of observations that can be used to build a probabilistic model.
Discussion:
The core details of HMMs are discussed in the next paper.
This paper primarily deals with the two phases of recognition using a HMM:
(1) Encoding: In this stage, strokes are converted to geometric primitives.
(2)Segmentation and Recogntion: We need to have identified object classes apriori, and constructed HMMs for them. Individual object recognition is done using forward chaining. To recognize in an overall scene,we end up with an optimization of problem of assigning models to subsequences such that there is maximum likelihood of some scene.
No comments:
Post a Comment